CapitalEfficiencyinConcentratedLiquidity

Concentrated liquidity was supposed to maximize capital efficiency on decentralized exchanges. Yet between a quarter and a third of it still does not sit where trades happen: it sits outside the active range and earns nothing.

• 29.5% mean out-of-range, 26 weeks, TVL-weighted.
• $542M idle, average week, across four protocols.
• ~85% underutilized, v3-family.
• ~$150M foregone fees per year, across all venues.

In Uniswap v2, every dollar spread evenly across all prices, so only about 1–5% of a pool ever sat within a few percent of it. Concentrated liquidity was supposed to maximize capital efficiency on decentralized exchanges by keeping capital near the current price, where trading happens. And it partly worked: v3 let each LP set a price range for their capital, and most of the market has moved that way. But a large share of it still earns nothing, and the inefficiency is more layered than v2's: capital can be actively used, in range but untouched by the day's trades, or out of range entirely.

To quantify it, we rebuilt every position in the top ~200 pools by activity on each of Uniswap v3, PancakeSwap v3, Aerodrome Slipstream and Uniswap v4, every week through the first half of 2026, across seven chains and about $1.84 billion of liquidity in an average week. Averaged over the 26 weeks, 29.5% of that capital sat outside its active fee-earning range, earning nothing. We call it idle, though out of range is not always the same as neglected. In an average week that came to about $542 million.

The figure is stable. Apart from a single early-February spike, when the out-of-range share touched 41%, it held inside a 25–35% band throughout. However we cut the data, a quarter to a third of the capital sits outside its active range. And it is not all capital briefly waiting to come back: more than a third of it had gone 90 days without a single adjustment.

• Out of range is the floor. About 30% of concentrated-liquidity capital sits fully outside its active range, earning nothing, the strictest measure of idle. Add the in-range capital that is never actually traded against, and roughly 85% is underused overall.
• A third of the idle capital is dormant. More than a third of the out-of-range capital, around $200 million, has sat untouched for over 90 days, not just briefly waiting to come back into range.
• Distance strands liquidity more than volatility. The idle share rises with how far the price travels over a week, but it's not affected much by frequent small moves: a violent week that round-trips leaves positions in range, while a quiet one-way drift pushes them out.
• Most idle capital sits in individual wallets. Automated managers and bots keep their positions in range; the idle is overwhelmingly individuals who set a range and walked away. On Base, contracts hold about half the Uniswap v3 capital but carry only a fraction of the idle, so individuals account for 82% of it.
• No design escapes it. Which protocol looks idlest flips depending on the pair, and Uniswap v4, the newest architecture, sits right where v3 does at about 30%. Every venue carries roughly the same idle share.

Out of range is the strictest, most mechanical measure: the price sits outside a position's band, so by the AMM's design that capital earns no fees and provides no depth, meaning it is not available to trade against, and does nothing to hold the price steady. We call it inactive, or idle. But it is only a floor, because being in range is not the same as working. A position counts as in range the moment the price is anywhere inside its band, even a band so wide that only a sliver near the price does anything. The extreme case is a full-range position, identical to a v2 deposit: in range at all times, with almost none of its capital active.

Underutilized capital, whether out of range or in range but too far from the price to be touched, is the broader metric 1inch's Aqua whitepaper measured. We rebuild that calculation and reach a similar figure: about 85% of the capital is underutilized, steady across all 26 weeks. Averaged over the same 26 weeks, the v3-family capital breaks down as 13.7% actively used, 56.9% in range but unswept, and 29.4% out of range.

The 85% figure is not special to concentrated liquidity. We ran the same measure on the constant-product venues it replaced: the 200 largest pools on Uniswap v2, PancakeSwap v2 and Aerodrome's constant-product AMM on Base, asking the same question we asked of the CLMMs: how much of the standing liquidity did the day's trading cross?

Almost none. Across the 26 weeks about 98.7% of the capital sat outside the day's traded band: 98.5% on Uniswap v2, 98.8% on PancakeSwap v2, 98.5% on Aerodrome basic. Every week landed between 98.2% and 99.2%, three constant-product venues on three chains converging on the same number. As expected from pools that spread capital across every price, the underutilized share is consistently higher here than on concentrated-liquidity venues (CLMMs).

Out-of-range capital is not automatically wasted. Some LPs place ranges off to one side on purpose, where they act as standing limit orders: as the price crosses the range the position converts to the other token, filling a trade without paying a trading fee. Others run bots that rotate positions constantly. So we checked when each out-of-range position was last touched by its owner, meaning any deposit or withdrawal on that range.

Averaged across the 26 weeks, 43.8% of out-of-range capital had been touched in the past 30 days, which looks like active positioning; another 19.5% was 30 to 90 days old. The remaining 36.7%, roughly $200 million, had gone untouched for more than 90 days. We call this dormant: a 90-day-untouched position may be passive or automated rather than forgotten, so all we can say is that it has not been repositioned.

Averaged over the period, the venues split sharply. On Uniswap v3, 44.5% of out-of-range capital is dormant by the 90-day test, the largest single block in the study, and only 36% of its idle was touched in the past 30 days. Aerodrome sits at the other end, about 20% dormant, with 58% of its idle touched within 30 days. PancakeSwap and Uniswap v4 fall in between, each around 27% dormant, though Pancake's idle is fresher (57% touched within 30 days against v4's 43%).

The same mechanical idleness has different behavior underneath: Uniswap v3's idle is set-and-forget, positions left in place as the price walked away, while the other venues' is younger and churned, bots re-placing positions that keep falling out of range.

A second way to qualify idleness: how far the price has moved from the active point. A position 1% below the market might come back this week; one 200% away hardly ever does. About 35% of all out-of-range capital sits within 5% of the market, the genuinely "just displaced" band. The rest is further out: roughly 43% more than 25% away, and 17% more than 100% away, where the market would have to double or halve. The venues differ here too: Uniswap v3's idle clusters at moderate distances (34% within 5%, 9% beyond 100%); Uniswap v4 is more polarized, carrying more both close to the price and far past it (41% within 5%, 14% beyond 100%), fitting its younger, higher-velocity book. Time and distance agree: much of the out-of-range capital is not coming back soon.

We grouped positions by size and found the idle rate is monotonic: the smaller the position, the more likely it sits outside the range, and the smallest are idle more often than not.

Large positions hold most of the capital, so idle dollars concentrate at the top: positions over $1 million hold about 47% of all idle capital, roughly $260 million, and those over $100k about 76%. The many sub-$1k positions are the large majority of the count but only 1.6% of the idle dollars. So the problem is concentrated in large, well-resourced positions that drifted out of range, with a long tail of small ones that go idle more often but barely move the total.

The same pattern holds across all four venues: in every one, the largest positions are the least likely to sit out of range.

Uniswap v4 keeps v3's concentrated-liquidity model, the tick-based price ranges this report measures, so an out-of-range position on v4 means exactly what it does on any other venue. What changed sits underneath the pricing. Pools no longer each live in their own contract; they share one, which makes them cheaper to deploy and route through but leaves the idle question untouched. The consequential change is hooks: a pool can attach custom code that runs at set moments around a swap or a liquidity change, used to set a dynamic fee or alter swap logic, and in principle to do something with the parked tokens themselves, lend them out or place them in an outside protocol to earn while they sit out of range. That last possibility is the one real caveat to treating v4's out-of-range capital as idle, and we return to it below.

With the pricing model unchanged, v4 is no better on idle. Its top 200 pools hold about $230 million at the end of June, and 30.5% is out of range on average, right where Uniswap v3 sits. Its idle is younger than v3's (27% dormant against 45%), consistent with a younger, higher-velocity LP base rather than caused by a structural improvement, and on size it runs the familiar gradient (55% idle sub-$1k down to 22% over $1M), with its largest positions holding about a third of its idle dollars.

Only about a tenth of v4 TVL sits behind any hook, so the hookless ~90% carries the out-of-range claim outright, and the hooked pools are in fact less idle (about 12% out of range against 33%). Looking at what each hooked pool does: every one is a swap-logic, dynamic-fee or liquidity-accounting hook that leaves the tokens in the pool. None is a rehypothecation or external-yield hook, moving idle tokens into lending markets such as Aave or Morpho so they keep earning while out of range. The largest hooked pool here, 58% of hooked TVL, runs an after-swap accounting hook and sits essentially fully in range. So hooks do not necessarily make idle capital any more efficient, and on v4 as it stands, out of range means earning nothing.

The two sections that follow, on who holds the idle capital and on the fees it forgoes, cover the v3 family only. Uniswap v4's architecture blocks both: its positions run through a different position manager, and its dynamic-fee pools cannot be priced by the fixed-tier method used for the others.

A fair objection to calling this capital idle or dormant is that some of it may be working another way. A position can be run by an automated manager that rebalances it, or staked in a farm or gauge that pays rewards whether or not it earns fees. Such a position can be out of range yet still earning for its owner elsewhere. So we traced each position to its owner: staked capital sits with the farm or gauge contract rather than the individual depositor, so this trace separates incentive-driven liquidity from the rest.

On Uniswap v3, the idle capital is almost entirely individually held. On Ethereum, wallets own 91% of the capital and 94 cents of every idle dollar; on Arbitrum, 78% of the capital and 92% of the idle. Base makes the split clearest: contracts hold about half the Uniswap v3 capital there, but they sit in range, only 6.5% out of range against roughly 30% for wallets. So even though contracts hold half the money, individuals carry 82% of the idle. Everywhere, the contract-held remainder (active managers, market-making bots, routers) sits far more reliably in range.

Both incentivized venues pay rewards to in-range liquidity. That pushes LPs to keep positions active, and it shows in Aerodrome, whose staked capital is the least idle anywhere, about 16% out of range. But the reward does not eliminate idle: PancakeSwap's staked capital runs markedly higher, especially on BNB, where most of its liquidity sits and where sharp moves push pools out of range faster than LPs restore them. Rewarding in-range liquidity holds idle down, but does not prevent it when prices move fast.

This connects to the dormancy split. On Uniswap, the idle is set-and-forget: individuals who placed a range, walked away, and seldom repositioned. On the incentivized venues, LPs have a stronger reason to keep repositioning, so positions that fall out of range are soon replaced rather than left to sit. The idleness looks identical on the surface; underneath, Uniswap's tends to be dormant while the incentivized venues' keeps turning over.

Over the period analyzed, each dollar of in-range capital in the Uniswap and PancakeSwap pools earned about 35 cents a year in fees, while out-of-range capital earned nothing. Applying that rate to the idle capital, it would have collected roughly $150 million a year: about $116 million on Uniswap and $25 million on PancakeSwap, computed exactly from their fixed onchain fee tiers, plus an estimated $6–12 million on Aerodrome, whose Slipstream fees are not a fixed tier and so can only be bounded.

Those fees come from trading volume, not from how much capital sits in a pool, and only in-range liquidity collects them. So moving the idle capital into range would not generate more fees; it would share the same fees among more liquidity and pull everyone's rate down. The $150 million then measures what each idle LP gave up on its own, as if it alone had moved into range while the rest stayed put. It is also why that rate looks high: a fixed pot of fees spread over a thin sliver of in-range capital makes that sliver earn a lot.

Uniswap v4 is left out here: its dynamic fees can't be priced by the fixed-tier method used for v3 and Pancake. See the methodology for detail.

Across the 26 weeks, the idle share across all pairs rose with the size of the weekly price move, with ETH's weekly return as the market-move reference: about 0.4 percentage points for every 1% ETH moved (r ≈ 0.49). That part is much as you would expect: when the price travels, positions fall out of range faster than LPs recenter them, and depth thins just when traders need it most. The idlest weeks were the two with the sharpest moves, an early-February spike and an early-June selloff, while through a calm spring the idle share barely responded to small moves.

The more counterintuitive result is that volatility on its own barely moves the idle share. What strands liquidity is how far the price ends up from where it started over the week, not how much it churned along the way. A violent week that round-trips back near its starting price leaves most positions in range; a quieter, one-way drift pushes them out. So realized volatility, the standard measure of how much a market moves around, hardly tracks the idle share at all: r ≈ 0.27 over 7 days, and essentially zero (−0.04) over 30.

We compared the same pairs across venues, and the ranking flips depending on the pair. Uniswap has the highest idle share on BTC/stable (41.9%) and ETH/BTC (35.1%), while on ETH/stable it is PancakeSwap that is higher (43.7%, against Uniswap's 32.8%). Aerodrome (Base only) reads lowest on all three (28.7%, 19.8%, 16.7%).

This is why a single idle number per protocol is misleading. What looks like a protocol gap is really a difference in what each venue happens to host, a bot-managed pool here, a different chain mix there. When pooling the pairs together the venues converge into a tight band in the high 20s to low 30s.

Even stablecoin pairs are no exception. In a pool like USDC/USDT the price barely moves, yet about 30% of the capital is still out of range, in line with the rest of the market and consistent across venues. The reason is that LPs concentrate stablecoin liquidity into very tight ranges, a few basis points wide, so even the small movement a peg allows is enough to push a position out. What matters is the size of the move relative to the width of the range, and for stablecoins both are tiny, so they net out to the same place as everything else.

The idle problem sits inside a fragmentation problem. Across the period, ETH/stablecoin volume trades over about 170 venues a week, and the largest never carries more than 40% of it (28.8% on average). BTC pairs are more concentrated: the top venue peaked at 57.8% for BTC/stablecoin as flow moved to Aerodrome on Base, and 52.5% for ETH/BTC. Even there, none holds 60%. An LP or trader on a single venue sees at best about 60% of the market, usually far less.

Idle capital and fragmentation are two inefficiencies on the same market: much of the capital is out of range, and even the capital in range is split across more venues than any single LP or router reaches. This report rebuilds that picture, but to act on the whole market at once, an LP or a router needs to see it as it happens, which pool holds the depth and where volume is clearing, across every venue. That live, consolidated view of DEX activity is what Dune is building for institutions.

The results here survive the obvious challenges: not an artifact of pool selection, steady across all 26 weeks, reconciled to onchain balances within a few percent, with the checks set out in the methodology. What they show is that concentrated liquidity was a real improvement, cutting the underutilized share from about 99% on the constant-product venues it replaced, and that capital efficiency on decentralized exchanges is still far from solved: a quarter to a third of the capital sits out of range earning nothing, most of it in individually held positions that drift out and are never moved back.

Three levers address it. The most direct is active management: automated managers and managed vaults that recenter as the price moves, whose positions in this data stay in range far more reliably than hand-set ones. The second puts idle capital to work while it waits, routing it through hooks or wrappers into lending markets so it keeps earning, which Uniswap v4 makes possible though almost none of the pools we measured yet use. The third is incentive design: rewarding in-range liquidity measurably lowers the idle share, as Aerodrome shows, without removing it. None is free: each trades gas, complexity, or contract risk against the yield it recovers, and which fits depends on the position.

For decentralized exchanges to compete with centralized venues and traditional trading, they need to attract more liquidity, and increasingly the institutional-size capital looking to add onchain liquidity provision to its mix of yield streams. That capital allocates where it can see the market clearly. Acting on the inefficiencies this report measures, when to recenter a range, which pool to enter, where to route an order, depends on a comprehensive, clean, and increasingly real-time view of where liquidity sits, how deep it is, and where the flow is, across every venue and chain. That is the real-time DEX market-data feed Dune is building for institutions.

We rebuild every position in the study from its onchain deposit and withdrawal history, for the roughly 200 most active pools on each of Uniswap v3, PancakeSwap v3, Aerodrome Slipstream and Uniswap v4, and value each one from the pool's state and daily prices. We take 26 weekly snapshots, one per week from January 6 to June 30 2026, each pinned to an exact block, giving about 6.5 million position-snapshots in a single combined dataset that every figure is drawn from. The position-level cuts (composition, dormancy, distance, size and the v4 hook split) are averaged across all 26 weeks; the one exception is ownership, reported at the June 30 snapshot because it is rebuilt separately.

The set of pools is chosen once and held fixed: we rank by the most recent month's activity and follow the same pools back to January, so the panel grows from 559 to 776 pools as the newer ones come into existence, a survivorship tilt we disclose rather than correct. Uniswap and PancakeSwap pools are ranked by dollar volume; Aerodrome Slipstream by number of swaps, because its dollar volume is not cleanly captured in the standard tables. Coverage by protocol:

• Uniswap v3: Ethereum, Base, Arbitrum, Unichain, Polygon and Optimism
• Uniswap v4: Ethereum, Base, Arbitrum, BNB, Unichain, Polygon, and Optimism
• PancakeSwap v3: BNB, Base, Arbitrum and Ethereum
• Aerodrome Slipstream: Base

Figures are weighted by the capital in each pool rather than by pool count, so a pool with fifty dollars in it does not count the same as one with fifty million. Weighted this way, the out-of-range share is about 30%; measured pool by pool, the median is lower, 14 to 23% depending on the venue, because the many small pools are individually less idle. That gap is itself one of the findings: idle capital concentrates in the large pools.

Every reconstructed position is checked against the tokens the pool actually holds onchain, and the two agree to within a few percent, about 97% by value across Uniswap, PancakeSwap and Aerodrome. The small gap is uncollected fees, which sit outside every range and so do not change the idle share. The agreement holds in the early weeks of the period as well as the recent ones, and tightens further at exact state blocks; weekly snapshots track daily ones within about a percentage point. Uniswap v4 is checked differently, because its shared-vault design means a pool's balances cannot be reconciled one by one; there the reconstructed active liquidity matches what each pool reports at the current price, which confirms the liquidity but leaves its dollar validation lighter than the rest. The headline is also robust to coverage: within every protocol the idle share is heaviest in the largest, most-traded pools and falls down the tail, so widening the universe would lower the figure, not raise it.

• Out of range is the strict measure: at each weekly snapshot, a position whose band does not contain the current price.
• Underutilized is broader, adding in-range capital that the week's trading never reached; because Uniswap v4's architecture does not allow the same clean split, the ~85% underutilized figure covers the v3 family, while v4 still enters the out-of-range, dormancy, distance and size cuts.
• Foregone fees are the average out-of-range capital multiplied by the fee rate in-range capital earned over the period (about 35%, annualized), taken from each pool's fee tier for Uniswap and PancakeSwap and bounded for Aerodrome, whose fees are dynamic; Uniswap v4 is excluded, since its dynamic fees cannot be priced this way.
• Pair classes are read from each pool's traded tokens; the stablecoin reading (about 30% out of range) covers pools of two major stablecoins and holds however the set of stablecoins is drawn. The volatility fit relates the whole panel's weekly out-of-range share to ETH's weekly move, with ETH standing in for the market rather than limiting the sample to ETH pairs; trailing 7- and 30-day volatility are tested the same way.
• Owners are found by following each position's liquidity NFT to whoever holds it now, with staked capital attributed to its farm or gauge contract. Two caveats matter for reading the result: a wallet-held position is individually held but not necessarily retail, since professional desks and bots also trade from ordinary wallets, so size is the better guide to scale; and because only some manager contracts are publicly labeled, the share held by managers is if anything understated.

As a baseline, we run the same measure on the three largest venues that predate concentrated liquidity (Uniswap v2, PancakeSwap v2 and Aerodrome's constant-product pools), where every dollar is spread across all prices by design. On these, about 98.7% of the capital sits outside the day's traded band, steady in a narrow 98.2 to 99.2% range week to week. That is the standard concentrated liquidity improved on, and it is why an 85% underutilized share still counts as progress.

The measured universe is EVM concentrated liquidity. Solana's concentrated-liquidity venues (Raydium, Orca, Meteora) are a large market we have left to a dedicated follow-up rather than fold in at a lower standard of validation. Curve is out of scope, because its LPs do not choose price ranges and so have no in-range or out-of-range state to measure.

Every trade in this report was read from a single normalized table. DEX.TRADES gives one schema across every DEX and chain: price, volume, tokens, and pool, decoded and reconciled to the source. It's the same feed behind the fragmentation, fee, and pair-level analysis here, and it's what powers the real-time DEX market-data feed Dune is building for institutions.